基于局部背景加权和能量约束的多目标检测与跟踪算法  

Multi-target Detection and Tracking Algorithm Based on Local Background Weighting and Energy Constraint

作　　者：陈家红[1] 

机构地区：[1]金陵科技学院计算机工程学院,南京211169

出　　处：《计算机工程》2015年第11期267-272,共6页Computer Engineering

基　　金：江苏省高校自然科学基金资助项目(11KJD520006);江苏省教育科学"十二五"规划课题基金资助项目(D/2013/01/068)

摘　　要：针对标记点处理方法用于多目标跟踪时效果不佳的问题,在标记点方法的基础上,提出一种多目标联合检测跟踪算法。改进的帧间差法用于目标的初步确定,通过局部背景加权进一步确定其与多个目标的标记。考虑动态模型的轨迹一致性问题,对动态目标、长时间跟踪和目标互斥相似等问题进行研究。针对非凸性的能量函数采用可逆跳转马尔可夫链蒙特卡洛进行优化。实验结果表明,在有高斯噪声情况下,与其他跟踪算法相比,该算法的检测和跟踪相似度最高,在卫星图像序列和自采集视频中的精度和召回率也最高,整体性能较优。As the performance of multi-target tracking based on Marked Point Process （MPP） is poor, a hybrid multitarget detection and tracking is proposed. The improved frame difference method is used to determine the preliminary of the target. Local background weighting is applied to identify further and mark the targets. The consistency of the dynamic model is considered. Energy is used to constraint target dynamics, long-term tracking and target similarity and mutual exclusion. Reversible jump Markov chain Monte Carlo is applied to optimize the nonconvex energy function. Some image sequences are added with different levels of noise. Experimental results show that in the image sequences with Gaussian noise ,compared with other tracking algorithms, the similarities of the proposed algorithm＇ s detection and tracking are highest. In addition, the precision and recall in satellite image sequences and self-capture videos are also highest. The overall performance of the proposed algorithm is better.

关 键 词：多目标跟踪 局部背景加权 能量约束 非凸性 可逆跳转马尔可夫链蒙特卡洛 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]